SemiAnalysis: China's CXMT Is Set to Challenge DRAM Incumbents#
| URL | https://newsletter.semianalysis.com/p/chinas-cxmt-is-set-to-challenge-dram |
| Published | 2026-06-23 |
| Read | 2026-09-26 (free section) |
| Paywall | The section on CXMT's equipment ecosystem under export controls, and China's wider wafer-fab-equipment outlook |
Summary (our words)#
CXMT's growth:
- Revenue: about $8.6B in 2025 (+156%), its first profitable year, and about $7.3B in Q1 2026 alone at about 70% operating margin.
- Wafer capacity: about 265k wafers/month at end 2025 → about 350k (2026) → 420k (2027, about 17% of global) → 500k (2028).
- Bit share: about 9% (2025) → about 12% (2027).
Its limits:
- Process: stuck at a 1z-equivalent node; commodity cost per bit is 30%+ above the incumbents.
- HBM: about 25% overall yield; HBM wafers rise from about 5k (2025) to about 100k (2028), about 12% of global HBM.
Claims#
| ID | Claim | Evidence |
|---|---|---|
| SA-cxmt-1 | Wafer capacity about 350k/month end 2026, 420k end 2027, 500k end 2028 | Free section |
| SA-cxmt-2 | Bit share about 9% (2025) → about 12% (2027) | Free section |
| SA-cxmt-3 | Commodity DRAM cost per bit 30%+ higher than competitors | Free section |
| SA-cxmt-4 | Q1 2026 operating margin about 70% | Free section |
| SA-cxmt-5 | HBM yield about 25% overall; HBM about 12% of global share by 2028 | Free section |
Our notes#
- This corrects our earlier view that CXMT was the most underrated reason consumer DDR5 could ease early. About +3 points of bit share over two years is relief at the margin, not a cycle-breaker.
- It is fungible, though: Chinese buyers who switch to CXMT free incumbent supply for everyone else.
- At current prices, the higher cost doesn't matter; CXMT sells everything at about 70% margin. It would matter only after prices fall.
- Wafers aren't bits. CXMT's older process gives fewer bits per wafer, so its share of wafer capacity (about 17%) overstates its share of supply (about 12%).